Overview

Machine Learning Engineer — Detection & Intervention Systems

Responsibilities

• Execution friction — clarification loops, stalled decisions, ownership gaps, blocked work.
• Emotional friction — disengagement, overload, tone shifts, boundary erosion.
• Build and maintain rule-based and NLP detection systems as the primary decision-making layer — fast, explainable, and cheap to run at scale.
• Design the logic that decides when a signal is confidently resolved by rules versus when it genuinely needs deeper model-based reasoning.
• Own confidence scoring and thresholds — the layer that decides whether the product is even allowed to act on what it detected.
• Implement and tune a tiered intervention model — from invisible observation through to escalation — including cooldown logic so the product never nags.
• Build and maintain the datasets and test suites that keep detection accurate as the product evolves, and close the feedback loop from real usage back into the system.
• Work directly across major workplace communication platforms and APIs (chat, email, calendar, docs, and adjacent tools).
• Ship inside a Python/FastAPI + PostgreSQL backend, with a clear line to the dashboard consuming your outputs.
• Sit close enough to the founder and the rest of the six-person team that architecture decisions get made in a conversation, not a design-doc review cycle.
• Strong Python engineering — production code, not notebooks. FastAPI or an equivalent async framework.
• Real experience with NLP: rule-based / pattern-based detection systems, not just calling a model API and hoping.
• Comfort designing deterministic systems — state machines, threshold logic, scoring/calibration — where correctness and explainability matter as much as accuracy.
• Working knowledge of PostgreSQL and building/maintaining datasets for training, validation, and regression testing.
• Judgment about when to use an LLM and when not to. You should be able to argue against using one.
• Strongly preferred

Qualifications

• Prior experience at an early-stage startup where you owned a system end to end, not a slice of one.
• Exposure to synthetic dataset generation for classification tasks.
• You'd rather ship a simpler rule that works than a clever model that's a black box — but you know exactly when that trade-off flips.
• You're comfortable with ambiguity moving fast — specs here evolve as we learn, and you adapt without needing everything locked down first.
• You push back with reasoning, not ego, and you don't need to be told twice once a decision is made.
• You care about precision in language and architecture as much as in code — if a term is ambiguous, you'll ask before you build.
• You want ownership, not a ticket queue. In a six-person team, whatever you build, you also watch in production.
• You're not looking for a place that has it all figured out. We don't. You're looking for the chance to figure it out and get real equity in the outcome.
• Can you commit to working on-site or hybrid out of Hyderabad? (Yes/No)
• Have you built a production rule-based or NLP detection system before, not just called a third-party LLM API? (Yes/No)
• Are you comfortable joining a 6-person, pre-revenue, pre-seed team with no guaranteed runway beyond the current raise? (Yes/No)
• Can you reason through a rule-based/deterministic approach before defaulting to "just use a model"? (Yes/No)
• Are you available to start within [X weeks] of an offer? (Yes/No)